langroid 0.2.5__tar.gz → 0.2.7__tar.gz

{langroid-0.2.5 → langroid-0.2.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langroid
-Version: 0.2.5
+Version: 0.2.7
 Summary: Harness LLMs with Multi-Agent Programming
 License: MIT
 Author: Prasad Chalasani
@@ -150,7 +150,7 @@ This Multi-Agent paradigm is inspired by the
 `Langroid` is a fresh take on LLM app-development, where considerable thought has gone 
 into simplifying the developer experience; it does not use `Langchain`.
 
-Companies are using/adapting Langroid in production. Here is a quote from one of them:
+📢 Companies are using/adapting Langroid in **production**. Here is a quote:
 
 >[Nullify](https://www.nullify.ai) uses AI Agents for secure software development. 
 > It finds, prioritizes and fixes vulnerabilities. We have internally adapted Langroid's multi-agent orchestration framework in production, after evaluating CrewAI, Autogen, LangChain, Langflow, etc. We found Langroid to be far superior to those frameworks in terms of ease of setup and flexibility. Langroid's Agent and Task abstractions are intuitive, well thought out, and provide a great developer  experience. We wanted the quickest way to get something in production. With other frameworks it would have taken us weeks, but with Langroid we got to good results in minutes. Highly recommended! <br> -- Jacky Wong, Head of AI at Nullify.
@@ -461,17 +461,17 @@ such as [ollama](https://github.com/ollama), [oobabooga](https://github.com/ooba
 - **Observability, Logging, Lineage:** Langroid generates detailed logs of multi-agent interactions and
   maintains provenance/lineage of messages, so that you can trace back
   the origin of a message.
-- **[Tools/Plugins/Function-calling](https://langroid.github.io/langroid/quick-start/chat-agent-tool/)**: Langroid supports OpenAI's recently
-  released [function calling](https://platform.openai.com/docs/guides/gpt/function-calling)
-  feature. In addition, Langroid has its own native equivalent, which we
-  call **tools** (also known as "plugins" in other contexts). Function
-  calling and tools have the same developer-facing interface, implemented
+- **[Tools/Plugins/Function-calling](https://langroid.github.io/langroid/quick-start/chat-agent-tool/)**:
+  Langroid supports OpenAI's [function calling](https://platform.openai.com/docs/guides/gpt/function-calling), as
+  well as an equivalent `ToolMessage` mechanism which works with
+  any LLM, not just OpenAI's.
+  Function calling and tools have the same developer-facing interface, implemented
   using [Pydantic](https://docs.pydantic.dev/latest/),
   which makes it very easy to define tools/functions and enable agents
   to use them. Benefits of using Pydantic are that you never have to write
   complex JSON specs for function calling, and when the LLM
   hallucinates malformed JSON, the Pydantic error message is sent back to
-  the LLM so it can fix it!
+  the LLM so it can fix it.
 
 --- 
 

{langroid-0.2.5 → langroid-0.2.7}/README.md

@@ -46,7 +46,7 @@ This Multi-Agent paradigm is inspired by the
 `Langroid` is a fresh take on LLM app-development, where considerable thought has gone 
 into simplifying the developer experience; it does not use `Langchain`.
 
-Companies are using/adapting Langroid in production. Here is a quote from one of them:
+📢 Companies are using/adapting Langroid in **production**. Here is a quote:
 
 >[Nullify](https://www.nullify.ai) uses AI Agents for secure software development. 
 > It finds, prioritizes and fixes vulnerabilities. We have internally adapted Langroid's multi-agent orchestration framework in production, after evaluating CrewAI, Autogen, LangChain, Langflow, etc. We found Langroid to be far superior to those frameworks in terms of ease of setup and flexibility. Langroid's Agent and Task abstractions are intuitive, well thought out, and provide a great developer  experience. We wanted the quickest way to get something in production. With other frameworks it would have taken us weeks, but with Langroid we got to good results in minutes. Highly recommended! <br> -- Jacky Wong, Head of AI at Nullify.
@@ -357,17 +357,17 @@ such as [ollama](https://github.com/ollama), [oobabooga](https://github.com/ooba
 - **Observability, Logging, Lineage:** Langroid generates detailed logs of multi-agent interactions and
   maintains provenance/lineage of messages, so that you can trace back
   the origin of a message.
-- **[Tools/Plugins/Function-calling](https://langroid.github.io/langroid/quick-start/chat-agent-tool/)**: Langroid supports OpenAI's recently
-  released [function calling](https://platform.openai.com/docs/guides/gpt/function-calling)
-  feature. In addition, Langroid has its own native equivalent, which we
-  call **tools** (also known as "plugins" in other contexts). Function
-  calling and tools have the same developer-facing interface, implemented
+- **[Tools/Plugins/Function-calling](https://langroid.github.io/langroid/quick-start/chat-agent-tool/)**:
+  Langroid supports OpenAI's [function calling](https://platform.openai.com/docs/guides/gpt/function-calling), as
+  well as an equivalent `ToolMessage` mechanism which works with
+  any LLM, not just OpenAI's.
+  Function calling and tools have the same developer-facing interface, implemented
   using [Pydantic](https://docs.pydantic.dev/latest/),
   which makes it very easy to define tools/functions and enable agents
   to use them. Benefits of using Pydantic are that you never have to write
   complex JSON specs for function calling, and when the LLM
   hallucinates malformed JSON, the Pydantic error message is sent back to
-  the LLM so it can fix it!
+  the LLM so it can fix it.
 
 --- 
 

{langroid-0.2.5 → langroid-0.2.7}/langroid/agent/openai_assistant.py

@@ -100,6 +100,10 @@ class OpenAIAssistant(ChatAgent):
         super().__init__(config)
         self.config: OpenAIAssistantConfig = config
         self.llm: OpenAIGPT = OpenAIGPT(self.config.llm)
+        assert (
+            self.llm.cache is not None
+        ), "OpenAIAssistant requires a cache to store Assistant and Thread ids"
+
         if not isinstance(self.llm.client, openai.OpenAI):
             raise ValueError("Client must be OpenAI")
         # handles for various entities and methods
@@ -235,19 +239,23 @@ class OpenAIAssistant(ChatAgent):
         """Try to retrieve cached thread_id associated with
         this user + machine + organization"""
         key = self._cache_thread_key()
+        if self.llm.cache is None:
+            return None
         return self.llm.cache.retrieve(key)
 
     @no_type_check
     def _cache_assistant_lookup(self) -> str | None:
         """Try to retrieve cached assistant_id associated with
         this user + machine + organization"""
+        if self.llm.cache is None:
+            return None
         key = self._cache_assistant_key()
         return self.llm.cache.retrieve(key)
 
     @no_type_check
     def _cache_messages_lookup(self) -> LLMResponse | None:
         """Try to retrieve cached response for the message-list-hash"""
-        if not settings.cache:
+        if not settings.cache or self.llm.cache is None:
             return None
         key = self._cache_messages_key()
         cached_dict = self.llm.cache.retrieve(key)
@@ -260,6 +268,8 @@ class OpenAIAssistant(ChatAgent):
         Cache the assistant_id, thread_id associated with
         this user + machine + organization
         """
+        if self.llm.cache is None:
+            return
         if self.thread is None or self.assistant is None:
             raise ValueError("Thread or Assistant is None")
         thread_key = self._cache_thread_key()
@@ -336,7 +346,8 @@ class OpenAIAssistant(ChatAgent):
                         Could not delete thread with id {cached}, ignoring. 
                         """
                     )
-                self.llm.cache.delete_keys([self._cache_thread_key()])
+                if self.llm.cache is not None:
+                    self.llm.cache.delete_keys([self._cache_thread_key()])
         if self.thread is None:
             if self.assistant is None:
                 raise ValueError("Assistant is None")
@@ -392,7 +403,8 @@ class OpenAIAssistant(ChatAgent):
                         Could not delete assistant with id {cached}, ignoring. 
                         """
                     )
-                self.llm.cache.delete_keys([self._cache_assistant_key()])
+                if self.llm.cache is not None:
+                    self.llm.cache.delete_keys([self._cache_assistant_key()])
         if self.assistant is None:
             self.assistant = self.client.beta.assistants.create(
                 name=self.config.name,
@@ -614,7 +626,8 @@ class OpenAIAssistant(ChatAgent):
             usage=None,  # TODO
             cached=False,  # TODO - revisit when able to insert Assistant responses
         )
-        self.llm.cache.store(key, result.dict())
+        if self.llm.cache is not None:
+            self.llm.cache.store(key, result.dict())
         return result
 
     def _parse_run_required_action(self) -> List[AssistantToolCall]:

{langroid-0.2.5 → langroid-0.2.7}/langroid/agent/task.py

@@ -41,7 +41,6 @@ from langroid.parsing.routing import parse_addressed_message
 from langroid.pydantic_v1 import BaseModel
 from langroid.utils.configuration import settings
 from langroid.utils.constants import (
-    AT,  # regex for start of an addressed recipient e.g. "@"
     DONE,
     NO_ANSWER,
     PASS,
@@ -74,6 +73,22 @@ class TaskConfig(BaseModel):
         inf_loop_wait_factor (int): wait this * cycle_len msgs before loop-check
         restart_subtask_run (bool): whether to restart *every* run of this task
             when run as a subtask.
+        addressing_prefix (str): "@"-like prefix an agent can use to address other
+            agents, or entities of the agent. E.g., if this is "@", the addressing
+            string would be "@Alice", or "@user", "@llm", "@agent", etc.
+            If this is an empty string, then addressing is disabled.
+            Default is empty string "".
+            CAUTION: this is a deprecated practice, since normal prompts
+            can accidentally contain such addressing prefixes, and will break
+            your runs. This could happen especially when your prompt/context
+            contains code, but of course could occur in normal text as well.
+            Instead, use the `RecipientTool` to have agents address other agents or
+            entities. If you do choose to use `addressing_prefix`, the recommended
+            setting is to use `langroid.utils.constants.AT`, which currently is "|@|".
+            Note that this setting does NOT affect the use of `constants.SEND_TO` --
+            this is always enabled since this is a critical way for responders to
+            indicate that the message should be sent to a specific entity/agent.
+            (Search for "SEND_TO" in the examples/ dir to see how this is used.)
     """
 
     inf_loop_cycle_len: int = 10
@@ -81,6 +96,7 @@ class TaskConfig(BaseModel):
     inf_loop_wait_factor: int = 5
     restart_as_subtask: bool = False
     logs_dir: str = "logs"
+    addressing_prefix: str = ""
 
 
 class Task:
@@ -211,7 +227,7 @@ class Task:
             set_parent_agent=noop_fn,
         )
         self.config = config
-        # how to behave as a sub-task; can be overriden by `add_sub_task()`
+        # how to behave as a sub-task; can be overridden by `add_sub_task()`
         self.config_sub_task = copy.deepcopy(config)
         # counts of distinct pending messages in history,
         # to help detect (exact) infinite loops
@@ -1190,7 +1206,10 @@ class Task:
         if result is None:
             return None
         # if result content starts with @name, set recipient to name
-        is_pass, recipient, content = parse_routing(result)
+        is_pass, recipient, content = parse_routing(
+            result,
+            addressing_prefix=self.config.addressing_prefix,
+        )
         if is_pass is None:  # no routing, i.e. neither PASS nor SEND
             return result
         if is_pass:
@@ -1648,6 +1667,7 @@ class Task:
 
 def parse_routing(
     msg: ChatDocument | str,
+    addressing_prefix: str = "",
 ) -> Tuple[bool | None, str | None, str | None]:
     """
     Parse routing instruction if any, of the form:
@@ -1656,6 +1676,8 @@ def parse_routing(
     @<recipient> <content> (send content to recipient)
     Args:
         msg (ChatDocument|str|None): message to parse
+        addressing_prefix (str): prefix to address other agents or entities,
+             (e.g. "@". See documentation of `TaskConfig` for details).
     Returns:
         Tuple[bool|None, str|None, str|None]:
             bool: true=PASS, false=SEND, or None if neither
@@ -1682,8 +1704,12 @@ def parse_routing(
         else:
             return False, addressee, content_to_send
     if (
-        AT in content
-        and (addressee_content := parse_addressed_message(content, AT))[0] is not None
+        addressing_prefix != ""
+        and addressing_prefix in content
+        and (addressee_content := parse_addressed_message(content, addressing_prefix))[
+            0
+        ]
+        is not None
     ):
         (addressee, content_to_send) = addressee_content
         # if no content then treat same as PASS_TO

{langroid-0.2.5 → langroid-0.2.7}/langroid/language_models/base.py

@@ -17,6 +17,7 @@ from typing import (
 )
 
 from langroid.cachedb.base import CacheDBConfig
+from langroid.cachedb.redis_cachedb import RedisCacheConfig
 from langroid.parsing.agent_chats import parse_message
 from langroid.parsing.parse_json import top_level_json_field
 from langroid.prompts.dialog import collate_chat_history
@@ -50,7 +51,7 @@ class LLMConfig(BaseSettings):
     # use chat model for completion? For OpenAI models, this MUST be set to True!
     use_chat_for_completion: bool = True
     stream: bool = True  # stream output from API?
-    cache_config: None | CacheDBConfig = None
+    cache_config: None | CacheDBConfig = RedisCacheConfig()
 
     # Dict of model -> (input/prompt cost, output/completion cost)
     chat_cost_per_1k_tokens: Tuple[float, float] = (0.0, 0.0)

{langroid-0.2.5 → langroid-0.2.7}/langroid/language_models/openai_gpt.py

@@ -478,8 +478,9 @@ class OpenAIGPT(LanguageModel):
                 timeout=Timeout(self.config.timeout),
             )
 
-        self.cache: CacheDB
-        if settings.cache_type == "momento":
+        self.cache: CacheDB | None = None
+        use_cache = self.config.cache_config is not None
+        if settings.cache_type == "momento" and use_cache:
             from langroid.cachedb.momento_cachedb import (
                 MomentoCache,
                 MomentoCacheConfig,
@@ -492,7 +493,7 @@ class OpenAIGPT(LanguageModel):
                 # switch to fresh momento config if needed
                 config.cache_config = MomentoCacheConfig()
             self.cache = MomentoCache(config.cache_config)
-        elif "redis" in settings.cache_type:
+        elif "redis" in settings.cache_type and use_cache:
             if config.cache_config is None or not isinstance(
                 config.cache_config,
                 RedisCacheConfig,
@@ -505,10 +506,10 @@ class OpenAIGPT(LanguageModel):
                 # force use of fake redis if global cache_type is "fakeredis"
                 config.cache_config.fake = True
             self.cache = RedisCache(config.cache_config)
-        else:
+        elif settings.cache_type != "none" and use_cache:
             raise ValueError(
                 f"Invalid cache type {settings.cache_type}. "
-                "Valid types are momento, redis, fakeredis"
+                "Valid types are momento, redis, fakeredis, none"
             )
 
         self.config._validate_litellm()
@@ -818,6 +819,8 @@ class OpenAIGPT(LanguageModel):
         )
 
     def _cache_store(self, k: str, v: Any) -> None:
+        if self.cache is None:
+            return
         try:
             self.cache.store(k, v)
         except Exception as e:
@@ -825,6 +828,8 @@ class OpenAIGPT(LanguageModel):
             pass
 
     def _cache_lookup(self, fn_name: str, **kwargs: Dict[str, Any]) -> Tuple[str, Any]:
+        if self.cache is None:
+            return "", None  # no cache, return empty key and None result
         # Use the kwargs as the cache key
         sorted_kwargs_str = str(sorted(kwargs.items()))
         raw_key = f"{fn_name}:{sorted_kwargs_str}"

{langroid-0.2.5 → langroid-0.2.7}/langroid/mytypes.py

@@ -22,11 +22,17 @@ class Entity(str, Enum):
     SYSTEM = "System"
 
     def __eq__(self, other: object) -> bool:
-        """Allow case-insensitive comparison with strings."""
+        """Allow case-insensitive equality (==) comparison with strings."""
+        if other is None:
+            return False
         if isinstance(other, str):
             return self.value.lower() == other.lower()
         return super().__eq__(other)
 
+    def __ne__(self, other: object) -> bool:
+        """Allow case-insensitive non-equality (!=) comparison with strings."""
+        return not self.__eq__(other)
+
     def __hash__(self) -> int:
         """Override this to ensure hashability of the enum,
         so it can be used sets and dictionary keys.

{langroid-0.2.5 → langroid-0.2.7}/langroid/utils/configuration.py

@@ -16,7 +16,7 @@ class Settings(BaseSettings):
     progress: bool = False  # show progress spinners/bars?
     stream: bool = True  # stream output?
     cache: bool = True  # use cache?
-    cache_type: Literal["redis", "fakeredis", "momento"] = "redis"  # cache type
+    cache_type: Literal["redis", "fakeredis", "momento", "none"] = "redis"  # cache type
     interactive: bool = True  # interactive mode?
     gpt3_5: bool = True  # use GPT-3.5?
     chat_model: str = ""  # language model name, e.g. litellm/ollama/llama2

{langroid-0.2.5 → langroid-0.2.7}/langroid/utils/constants.py

@@ -18,6 +18,11 @@ DONE = "DONE"
 USER_QUIT_STRINGS = ["q", "x", "quit", "exit", "bye", DONE]
 PASS = "__PASS__"
 PASS_TO = PASS + ":"
-SEND_TO = "SEND:"
+SEND_TO = "__SEND__:"
 TOOL = "TOOL"
-AT = "@"
+# This is a recommended setting for TaskConfig.addressing_prefix if using it at all;
+# prefer to use `RecipientTool` to allow agents addressing others.
+# Caution the AT string should NOT contain any 'word' characters, i.e.
+# it no letters, digits or underscores.
+# See tests/main/test_msg_routing for example usage
+AT = "|@|"

{langroid-0.2.5 → langroid-0.2.7}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "langroid"
-version = "0.2.5"
+version = "0.2.7"
 description = "Harness LLMs with Multi-Agent Programming"
 authors = ["Prasad Chalasani <pchalasani@gmail.com>"]
 readme = "README.md"